Google is rolling out its next generation of Gemini models, with a focus on enhanced capabilities and affordability. The Gemini 2.0 family includes updates to Flash, the introduction of Flash-Lite, and an experimental version of Gemini 2.0 Pro, each designed for specific use cases and performance needs.

Gemini 2.0 Pro, in its experimental form, is positioned as Google's most capable model for coding and handling complex prompts. It boasts a 2 million token context window, allowing it to analyze and understand vast amounts of information. This model also has the ability to integrate external tools like Google Search and code execution.

Gemini 2.0 Flash, initially launched as an experiment, is now generally available. This model is designed for high-efficiency AI applications, providing low-latency responses and supporting large-scale multimodal reasoning. It features a 1 million token context window.

The newest addition, Gemini 2.0 Flash-Lite, aims to provide a cost-effective AI solution. Google claims it outperforms the previous 1.5 Flash model on most benchmarks while maintaining the same pricing and speed. Like the standard Flash model, it supports multimodal input and features a 1 million token context window. The company says it can generate a relevant one-line caption for around 40,000 unique photos for under a dollar in Google AI Studio’s paid tier.

This table compares the capabilities of different versions of Gemini, including 1.5 Flash, 1.5 Pro, 2.0 Flash-Lite, 2.0 Flash, and 2.0 Pro, across various benchmarks. It shows the performance of each version on tasks like general knowledge, code generation, reasoning, factuality, multilingual understanding, math, long-context understanding, image understanding, audio translation, and video analysis.

For users wanting to see the models "thinking," Gemini 2.0 Flash Thinking Experimental is now available to test in the Gemini app for free. This model breaks down prompts into a series of steps to strengthen its reasoning capabilities and show how it arrives at an answer.

The Gemini app is also getting a version of Flash Thinking Experimental that connects to Google Maps, YouTube, and Search. This integration allows the model to leverage external information and services to provide more comprehensive and relevant responses. For example, it can use Google Maps to calculate travel times or search YouTube for relevant videos.

Google is also emphasizing the safety and security of the Gemini 2.0 models. The company is using reinforcement learning techniques, where the AI critiques its own responses to improve accuracy. Automated red teaming is also used to identify vulnerabilities, including indirect prompt injection attacks.

Gemini 2.0 Flash-Lite is priced at $0.075 per million tokens (input) and $0.30 per million tokens (output).

With these updates, Google is expanding the availability and capabilities of its Gemini AI models, offering a range of options for developers and users with varying needs and budgets. It's a move that positions Google to compete with other AI companies, including DeepSeek, that are focusing on cost-efficient AI solutions.